CN116108181A - Customer information processing method, device and electronic equipment - Google Patents

Abstract

The application discloses a processing method and device of customer information and electronic equipment, wherein the method is applied to the field of big data and comprises the following steps: coding the client information in a plurality of incoming call lists of a plurality of clients to obtain a plurality of documents corresponding to the incoming call lists; acquiring a weighting matrix model of each document in a plurality of documents; acquiring a subject word set of each document in a plurality of documents; calculating the target similarity between each document and other documents in the plurality of documents according to the subject word set of each document and the weighted matrix model of each document; clustering a plurality of documents based on a PAM clustering algorithm according to the target similarity to obtain a clustered document set; and acquiring target information of a plurality of clients from the clustered document set. According to the method and the device, the problem that the information extracted from the electrician bill is inaccurate due to the fact that the value information of the electrician bill is extracted by adopting manual labeling or natural language processing technology in the related technology is solved.

Description

Customer information processing method, device and electronic equipment

technical field

The present application relates to the field of big data, and in particular, relates to a method, device and electronic equipment for processing customer information.

Background technique

When extracting the potential value information in multiple incoming work orders of multiple customers, manual classification and manual labeling are often used to extract information. The subjective influence of , resulting in poor quality and accuracy of labeling results, cannot objectively reflect the potential value information in multiple call work orders.

At present, natural language processing technology is also used to calculate text similarity, extract potential value information in multiple call orders, and convert unstructured text into structured information that is easy for computer recognition and processing, so as to realize the recognition of text information. dig. Traditional text similarity calculation models can be divided into three categories: Vector Space Model (VectorSpace Model, VSM), Generalized Vector Space Model (Generalized Vector Space Model, GVSM), and Latent Semantic Indexing (LSI) or It is called Latent Semantic Analysis (LSA). These three models generally use feature selection functions to extract keywords as text features when extracting features, such as word contribution TC, TFIDF, information entropy/information gain, mutual information, CHI statistics and other methods. However, traditional text similarity models require large-scale corpora and often ignore the grammatical and organizational structure as well as semantic information in the text. For example, the vector space model uses the bag-of-words model (Bag-of-words) to construct the feature space, and this model usually uses the "hard matching" method in feature matching, which cannot solve the problems of "multiple words with one meaning" and "multiple words with one word". righteousness" issue.

Aiming at the problem of inaccurate information extracted from the incoming work order by using manual annotation or natural language processing technology to extract the value information of the incoming work order in related technologies, no effective solution has been proposed so far.

Contents of the invention

The main purpose of this application is to provide a customer information processing method, device, and electronic equipment to solve the problem of extracting the value information of the incoming work order by manual labeling or natural language processing technology in related technologies, resulting in the inaccurate information extracted from the incoming work order. exact question.

In order to achieve the above object, according to one aspect of the present application, a method for processing customer information is provided. The method includes: encoding the customer information in multiple incoming work orders of multiple customers, and obtaining the multiple incoming work orders. A plurality of documents corresponding to the electrical work order, wherein the encoding formats of the plurality of documents are the same; obtaining a weighted matrix model of each of the plurality of documents; obtaining a set of subject words of each of the plurality of documents; According to the keyword set of each document and the weighted matrix model of each document, calculate the target similarity between each document in the plurality of documents and each other document; according to the target similarity, Clustering the multiple documents based on the PAM clustering algorithm to obtain a clustered document set; acquiring target information of multiple customers from the clustered document set.

Further, obtaining the weighted matrix model of each of the plurality of documents includes: deleting the preset character strings in the plurality of documents, and performing word segmentation and filtering on each of the plurality of documents, to obtain The first corpus; the first corpus is substituted into the word2vec model for calculation to obtain the word vector of each document; the plurality of documents are substituted into the TextRank model for calculation to obtain the keyword set of each document; according to each The similarity between the keyword set of the document and the keyword set of each other document, use the greedy selection algorithm to sort the keywords in the keyword set of each document, and obtain the keyword vector of each document; use the word The vector replaces the keywords in the keyword vector, and uses the TF-IDF feature selection function to calculate the weight of the word vector in the keyword vector to obtain the replaced keyword vector and the replaced keyword vector The word vector weight corresponding to the word vector in ; according to the replaced keyword vector and the word vector weight, the weighted matrix model of each document is obtained.

Further, substituting the multiple documents into the TextRank model for calculation, and obtaining the keyword set of each document includes: performing a first process on each of the multiple documents, and obtaining a word with a preset part of speech as each Candidate keywords of documents, wherein the first processing includes at least the following processing: sentence segmentation processing, word segmentation processing, filtering processing and part-of-speech tagging processing; according to each document and the candidate keywords of each document, Obtain the second corpus; use the TextRank model to convert the candidate keywords of each document into a keyword directed graph, and calculate the weight of the candidate keywords in each document to obtain the weighted keywords of each document A directed graph: from the weighted keyword directed graph, obtain candidate keywords whose weight is higher than a preset weight, and obtain the keyword set of each document.

其中，d代表预设的阻尼系数，a表示在所述加权关键词有向图中的关键词a，Out(b)表示在所述加权关键词有向图中关键词a指向的所有关键词的集合，In(a)表示在所述加权关键词有向图中所有指向关键词a的关键词的集合，b表示指向关键词a的关键词b，c表示关键词a指向的关键词c，S(b)表示关键词b的词影响力，tfidf_a表示关键词a的词频和逆文档频率的乘积，tfidf_c表示关键词c的词频和逆文档频率的乘积，w_ba表示在所述加权关键词有向图中关键词a和关键词b的权值，w_bc表示在所述加权关键词有向图中关键词b和关键词c的权值；对所述多个文档中的每个文档进行第二处理，获取第三语料库以及所述第三语料库中的主题词，其中，所述第二处理至少包括以下处理：分词处理、过滤处理、提取初始主题词处理和统计词频处理；依据所述词影响力得分的计算公式，迭代计算所述第三语料库中的主题词的词影响力得分，当第N次迭代计算中与第N-1次迭代的差值小于预设阈值时停止迭代计算，从第N次迭代计算的计算结果中，获取所述词影响力得分高于预设词影响力得分的所述主题词，得到所述每个文档的主题词集合。Further, obtaining the subject word set of each document in the plurality of documents includes: configuring the calculation formula of the word influence score of the word vector according to the TextRank model, and the calculation formula is as follows:

Wherein, d represents the preset damping coefficient, a represents keyword a in the weighted keyword directed graph, and Out(b) represents all keywords pointed to by keyword a in the weighted keyword directed graph In(a) represents the set of all keywords pointing to keyword a in the weighted keyword directed graph, b represents keyword b pointing to keyword a, and c represents keyword c pointing to keyword a , S(b) represents the word influence of keyword b, tfidf _a represents the product of word frequency and inverse document frequency of keyword a, tfidf _c represents the product of word frequency and inverse document frequency of keyword c, w _ba represents The weight value of keyword a and keyword b in the weighted keyword directed graph, w _bc represents the weight value of keyword b and keyword c in the weighted keyword directed graph; Each document is subjected to the second processing to obtain the third corpus and the subject words in the third corpus, wherein the second process includes at least the following processing: word segmentation processing, filtering processing, extraction of initial subject words processing and statistical word frequency processing ; According to the calculation formula of the word influence score, iteratively calculate the word influence score of the subject words in the third corpus, when the difference between the Nth iteration and the N-1 iteration is less than the preset threshold stop the iterative calculation, and obtain the subject words whose word influence score is higher than the preset word influence score from the calculation result of the Nth iterative calculation, and obtain the subject subject set of each document.

Further, according to the keyword set of each document and the weighted matrix model of each document, calculating the target similarity between each document in the plurality of documents and each other document includes: according to the The subject term set of each document constructs a bipartite graph model, to calculate the first similarity between the subject term set of each document and the subject term set of each other document; according to the least squares distance formula of the matrix, Calculate the second similarity between the weighted matrix model of each document and the weighted matrix model of each other document; calculate each document and each other according to the first similarity and the second similarity Target similarity between documents.

Further, constructing a bipartite graph model according to the subject term set of each document, so as to calculate the first similarity between the subject term set of each document and each other document subject term set includes: according to the Describe the subject word set of each document, construct the bipartite graph model of each document and each other document; use the Hungarian algorithm to calculate the bipartite matching maximum weight between each document and each other document, and combine all The maximum weight of the bipartite matching is used as the first similarity between the keyword set of each document and the keyword set of each other document.

Further, obtaining the target information of multiple customers from the clustered document collection includes: processing each clustered document collection according to the method of obtaining the subject word collection, and obtaining each clustered document collection The keyword set of the set; according to the method of obtaining the keyword set, process each clustered document set to obtain the keyword set of each clustered document set; according to the method of each clustered document set The keyword set and the keyword set of each clustered document set determine the target information of the plurality of customers.

Further, after calculating the target similarity between each document in the multiple documents and each other document according to the keyword set of each document and the weighted matrix model of each document, the The method further includes: adding the target similarity between each document in the plurality of documents and each other document to obtain a third similarity of each document; from the plurality of documents, Acquiring the target documents whose third similarity is higher than the preset similarity to obtain a target document set; pushing the target document set to the target object.

In order to achieve the above object, according to another aspect of the present application, a customer information processing device is provided, the device includes: a first acquisition unit, used to encode the customer information in multiple incoming work orders of multiple customers Processing to obtain a plurality of documents corresponding to the plurality of incoming work orders, wherein the encoding formats of the plurality of documents are the same; the second acquisition unit is configured to acquire a weighted matrix model of each document in the plurality of documents; The third obtaining unit is used to obtain the subject heading set of each document in the plurality of documents; the calculation unit is used to calculate the subject heading set of each document according to the weighting matrix model of each document. The target similarity between each document in the plurality of documents and each other document; the fourth acquisition unit is used to perform clustering on the plurality of documents based on the PAM clustering algorithm according to the target similarity, to obtain A clustered document set; a fifth acquiring unit, configured to acquire multiple customer target information from the clustered document set.

Further, the second acquisition unit includes: a first processing subunit, configured to delete preset character strings in the plurality of documents, and perform word segmentation and filtering processing on each of the plurality of documents, to obtain The first corpus; the first calculation subunit is used to substitute the first corpus into the word2vec model for calculation to obtain the word vector of each document; the second calculation subunit is used to substitute the multiple documents into the TextRank model Calculate in to obtain the keyword set of each document; the first acquisition subunit is used to use the greedy selection algorithm to divide each document according to the similarity between the keyword set of each document and each other Sort the keywords in the keyword set to obtain the keyword vector of each document; the third calculation subunit is used to replace the keywords in the keyword vector with the word vector, and use the TF-IDF feature The selection function calculates the weight of the word vector in the keyword vector, obtains the word vector weight corresponding to the word vector in the keyword vector after the replacement and the word vector in the keyword vector after the replacement; the second acquisition subunit uses and obtaining a weighted matrix model of each document according to the replaced keyword vector and the word vector weight.

Further, the second calculation subunit includes: a processing module, configured to perform first processing on each of the plurality of documents, and obtain a word with a preset part of speech as a candidate keyword for each document, wherein, The first processing includes at least the following processing: sentence segmentation processing, word segmentation processing, filtering processing, and part-of-speech tagging processing; the first acquisition module is used to acquire the second document according to each document and the candidate keywords of each document. The second corpus; the first calculation module, for using the TextRank model to convert the candidate keywords of each document into a keyword directed graph, and calculate the weight of the candidate keywords in each document, and obtain each The weighted keyword directed graph of the document; the second acquisition module is used to obtain the candidate keywords whose weight is higher than the preset weight from the weighted keyword directed graph, and obtain the weight of each document collection of keywords.

其中，d代表预设的阻尼系数，a表示在所述加权关键词有向图中的关键词a，Out(b)表示在所述加权关键词有向图中关键词a指向的所有关键词的集合，In(a)表示在所述加权关键词有向图中所有指向关键词a的关键词的集合，b表示指向关键词a的关键词b，c表示关键词a指向的关键词c，S(b)表示关键词b的词影响力，tfidf_a表示关键词a的词频和逆文档频率的乘积，tfidf_c表示关键词c的词频和逆文档频率的乘积，w_ba表示在所述加权关键词有向图中关键词a和关键词b的权值，w_bc表示在所述加权关键词有向图中关键词b和关键词c的权值；第二处理子单元，用于对所述多个文档中的每个文档进行第二处理，获取第三语料库以及所述第三语料库中的主题词，其中，所述第二处理至少包括以下处理：分词处理、过滤处理、提取初始主题词处理和统计词频处理；第四计算子单元，用于依据所述词影响力得分的计算公式，迭代计算所述第三语料库中的主题词的词影响力得分，当第N次迭代计算中与第N-1次迭代的差值小于预设阈值时停止迭代计算，从第N次迭代计算的计算结果中，获取所述词影响力得分高于预设词影响力得分的所述主题词，得到所述每个文档的主题词集合。Further, the third acquisition unit includes: a configuration subunit, configured to configure the calculation formula of the word influence score of the word vector according to the TextRank model, and the calculation formula is as follows:

Wherein, d represents the preset damping coefficient, a represents keyword a in the weighted keyword directed graph, and Out(b) represents all keywords pointed to by keyword a in the weighted keyword directed graph In(a) represents the set of all keywords pointing to keyword a in the weighted keyword directed graph, b represents keyword b pointing to keyword a, and c represents keyword c pointing to keyword a , S(b) represents the word influence of keyword b, tfidf _a represents the product of word frequency and inverse document frequency of keyword a, tfidf _c represents the product of word frequency and inverse document frequency of keyword c, w _ba represents The weight value of keyword a and keyword b in the weighted keyword directed graph, w _bc represents the weight value of keyword b and keyword c in the weighted keyword directed graph; the second processing subunit is used for Performing a second process on each of the plurality of documents to obtain a third corpus and subject words in the third corpus, wherein the second process at least includes the following processes: word segmentation processing, filtering processing, extraction Initial topic word processing and statistical word frequency processing; the fourth calculation subunit is used to iteratively calculate the word influence score of the topic words in the third corpus according to the calculation formula of the word influence score, when the Nth iteration When the difference between the calculation and the N-1th iteration is less than the preset threshold, the iterative calculation is stopped, and the word influence score higher than the preset word influence score is obtained from the calculation result of the Nth iteration calculation. subject words, to obtain a set of subject words of each document.

Further, the calculation unit includes: a construction subunit, configured to construct a bipartite graph model according to the keyword set of each document, so as to calculate the difference between the keyword set of each document and the keyword set of each other document. The first similarity between the collections; the fifth calculation subunit is used to calculate the second similarity between the weighted matrix model of each document and the weighted matrix model of each other document according to the least squares distance formula of the matrix Similarity: a sixth calculation subunit, configured to calculate a target similarity between each document and each other document according to the first similarity and the second similarity.

Further, the first construction subunit includes: a construction module, configured to construct the bipartite graph model of each document and each other document according to the keyword set of each document; a second calculation module, It is used to calculate the maximum weight of bipartite matching between each document and each other document using the Hungarian algorithm, and the maximum weight of bipartite matching is used as the set of subject words of each document and the set of subject words of each other document The first similarity between.

Further, the fifth acquisition unit includes: a third processing subunit, configured to process each clustered document set according to the method of acquiring the subject word set, and acquire the subject of each clustered document set word set; the fourth processing subunit is used to process each clustered document set according to the method for obtaining the keyword set, and obtain the keyword set of each clustered document set; determine the subunit, use The target information of the plurality of customers is determined according to the keyword set of each clustered document set and the keyword set of each clustered document set.

Further, the device further includes: a sixth acquisition unit, configured to calculate the target similarity between each document in the plurality of documents and each other document according to the subject word set and the weighted matrix model After the degree of similarity, add the target similarity between each of the multiple documents and each of the other documents to obtain the third similarity of each document; the seventh acquisition unit is used to obtain from the Among the plurality of documents, the target document whose third similarity is higher than the preset similarity is obtained to obtain a target document set; a push unit is used to push the target document set to the target object

In order to achieve the above object, according to one aspect of the present application, an electronic device is provided, including one or more processors and a memory, and the memory is used to store one or more programs, wherein, when one or more programs are used by one or more When executed by multiple processors, one or more processors can implement any one of the above-mentioned customer information processing methods.

Through this application, the following steps are adopted: encode the customer information in multiple incoming work orders of multiple customers, and obtain multiple documents corresponding to the multiple incoming work orders, wherein the encoding format of the multiple documents Same; Obtain the weighted matrix model of each document in the plurality of documents; Obtain the set of subject terms of each document in the plurality of documents; According to the set of subject terms of each document and the weighting of each document A matrix model, calculating the target similarity between each document in the plurality of documents and each other document; according to the target similarity, clustering the plurality of documents based on the PAM clustering algorithm to obtain a cluster After the collection of documents; the target information of multiple customers is obtained from the document collection after the clustering, which solves the problem of extracting the value information of the call work order by manual labeling or natural language processing technology, which leads to the inconsistency of the information extracted from the call work order. exact question. By calculating the weighted matrix model of each document and the subject word set of each document in multiple incoming call tickets, to calculate the target similarity between multiple documents, so as to extract target information in multiple documents, the algorithm is implemented Automatically extract target information from multiple documents, avoiding the impact of manual labeling of multiple documents on the information quality of target information, and achieve the effect of extracting more accurate target information from multiple call work orders.

Description of drawings

The drawings constituting a part of the application are used to provide further understanding of the application, and the schematic embodiments and descriptions of the application are used to explain the application, and do not constitute an improper limitation to the application. In the attached picture:

FIG. 1 is a flow chart of a method for processing customer information provided according to an embodiment of the present application;

FIG. 2 is a first schematic diagram of an optional customer information processing method provided according to an embodiment of the present application;

FIG. 3 is a second schematic diagram of an optional customer information processing method provided according to an embodiment of the present application;

FIG. 4 is a third schematic diagram of an optional customer information processing method provided according to an embodiment of the present application;

Fig. 5 is a schematic diagram of a processing device for customer information provided according to an embodiment of the present application;

Fig. 6 is a schematic diagram of an electronic device for processing customer information provided according to an embodiment of the present application.

Detailed ways

It should be noted that, in the case of no conflict, the embodiments in the present application and the features in the embodiments can be combined with each other. The present application will be described in detail below with reference to the accompanying drawings and embodiments.

In order to enable those skilled in the art to better understand the solution of the present application, the technical solution in the embodiment of the application will be clearly and completely described below in conjunction with the accompanying drawings in the embodiment of the application. Obviously, the described embodiment is only It is an embodiment of a part of the application, but not all of the embodiments. Based on the embodiments in this application, all other embodiments obtained by persons of ordinary skill in the art without creative efforts shall fall within the scope of protection of this application.

It should be noted that the terms "first" and "second" in the description and claims of the present application and the above drawings are used to distinguish similar objects, but not necessarily used to describe a specific sequence or sequence. It should be understood that the data so used may be interchanged under appropriate circumstances for the embodiments of the application described herein. Furthermore, the terms "comprising" and "having", as well as any variations thereof, are intended to cover a non-exclusive inclusion, for example, a process, method, system, product or device comprising a sequence of steps or elements is not necessarily limited to the expressly listed instead, may include other steps or elements not explicitly listed or inherent to the process, method, product or apparatus.

It should be noted that the user information (including but not limited to user equipment information, user personal information, etc.) and data (including but not limited to data used for analysis, stored data, displayed data, and customer Electrician orders, etc.), are all information and data authorized by the user or fully authorized by all parties, and the collection, use and processing of relevant data must comply with relevant laws, regulations and standards of relevant countries and regions, and provide corresponding operation entrances , for the user to choose to authorize or deny.

The present invention will be described below in conjunction with preferred implementation steps. Fig. 1 is a flow chart of the processing method of customer information provided according to the embodiment of the present application. As shown in Fig. 1, the method includes the following steps:

Step S101 , encoding customer information in multiple incoming call work orders of multiple customers to obtain multiple documents corresponding to the multiple incoming call work orders, wherein the encoding formats of the multiple documents are the same.

At present, the multiple call orders of multiple customers are generally multiple text files that record the communication process between the customer and the staff. In this embodiment, in order to further process the customer information, the multiple text files need to be converted Multiple documents in the same encoding format.

Step S102, acquiring a weighted matrix model of each of the multiple documents.

Step S103, obtaining a set of subject words of each of the multiple documents.

Step S104, according to the keyword set of each document and the weighted matrix model of each document, calculate the target similarity between each document and each other document among the plurality of documents.

In this embodiment, in order to calculate the target similarity between each document in multiple documents and each other document, after obtaining multiple documents with the same encoding format, it is necessary to calculate multiple documents to obtain multiple documents A weighted matrix model for each document in , and a collection of topic terms for each of the multiple documents.

Step S105, according to the target similarity, cluster the multiple documents based on the PAM clustering algorithm to obtain a clustered document set.

Step S106, acquiring target information of multiple customers from the clustered document collection.

In the prior art, when analyzing multiple incoming work orders of multiple customers and extracting information that staff are interested in, manual labeling is generally used to mark the multiple incoming work orders of multiple customers. However, the work efficiency of manually labeling multiple incoming work orders is low, and it cannot meet the current needs of big data analysis. Moreover, due to the subjective differences of different annotators, there is no uniform standard for the marking effect. The information extracted from multiple incoming work orders of poor quality.

In this embodiment, in order to improve the work efficiency of labeling multiple call work orders and improve the quality of information extracted from multiple call work orders, it is necessary to cluster multiple documents through the PAM clustering algorithm to obtain the clustered The clustered document collection, and then obtain the target information of multiple customers from the clustered document collection. Among them, the target information includes multiple customers’ experience feedback on the company’s products and services, and the target information also includes the potential business handling needs of customers. Extracting information from multiple incoming calls from multiple customers can improve customer service and accuracy. Marketing provides favorable support.

To sum up, the method for processing customer information provided by the embodiment of the present application obtains multiple documents corresponding to multiple incoming work orders by encoding the customer information in multiple incoming work orders of multiple customers, wherein, The encoding format of multiple documents is the same; obtain the weighted matrix model of each document in multiple documents; obtain the set of subject terms of each document in multiple documents; according to the set of subject terms of each document and the weighted matrix model of each document , to calculate the target similarity between each document in multiple documents and each other document; according to the target similarity, multiple documents are clustered based on the PAM clustering algorithm to obtain a clustered document set; from the clustering The target information of multiple customers is obtained from the final document collection, and the problem of inaccurate information extracted from the incoming work order is solved by using manual annotation or natural language processing technology to extract the value information of the incoming work order. By calculating the weighted matrix model of each document and the subject word set of each document in multiple incoming call tickets, to calculate the target similarity between multiple documents, so as to extract target information in multiple documents, the algorithm is implemented Automatically extract target information from multiple documents, avoiding the impact of manual labeling of multiple documents on the information quality of target information, and achieve the effect of extracting more accurate target information from multiple call work orders.

Optionally, in the method for processing customer information provided in the embodiment of the present application, obtaining the weighted matrix model of each of the multiple documents includes: deleting the preset character strings in the multiple documents, and Segment and filter each document to obtain the first corpus; Substitute the first corpus into the word2vec model for calculation to obtain the word vector of each document; Substitute multiple documents into the TextRank model for calculation to obtain the keywords of each document Set; according to the similarity between the keyword set of each document and the keyword set of each other document, use the greedy selection algorithm to sort the keywords in the keyword set of each document, and obtain the keyword vector of each document ;Use the word vector to replace the keywords in the keyword vector, and use the TF-IDF feature selection function to calculate the weight of the word vector in the keyword vector, and obtain the replaced keyword vector and the words in the replaced keyword vector The word vector weight corresponding to the vector; according to the replaced keyword vector and word vector weight, the weighted matrix model of each document is obtained.

At present, in the process of extracting the core content from the text through the natural language analysis technology, a small set of discrete categories is generally extracted from the text or the word vector corresponding to the text is calculated using the word2vec model. Using a small set of discrete categories to process text, that is, using noun phrases and verb phrases in the text to represent the text, however, the method of processing text through discrete small sets cannot fully capture the richness of multiple phrases in the text, and requires a huge feature space. Comparing the method of calculating word vectors using the word2vec model with the previous technology based on different types of neural networks with the best performance, it is found that using the word2vec model has a greater improvement in accuracy at a lower computational cost. However, using the word vector representation An inherent limitation of text is that word vectors do not care about the order of words in the text, and word vectors cannot represent phrases that conform to language habits. For example, word vectors cannot accurately represent the meaning expressed by idioms. In addition, the existing use of word vectors Models that represent text, for example, word vector mean model, word vector clustering model, and doc2vec model, the above-mentioned models that use word vectors to represent text do not consider the influence of words in the text.

In this embodiment, in order to obtain the meaning that can more accurately represent the text, it is first necessary to perform word segmentation and filtering on multiple documents to obtain the first corpus for subsequent calculations; then substitute the first corpus into the word2vec model for Calculate to obtain the word vector corresponding to each document; then substitute multiple documents into the TextRank model for calculation to obtain the keyword set of each document; then calculate the relationship between each keyword in the keyword set corresponding to each document and every other The similarity of each keyword in the keyword set corresponding to each document, according to the size of the similarity value between keywords, use the greedy selection algorithm to sort at least one keyword in the keyword set corresponding to each document, Obtain the keyword vector corresponding to each document; use the word vector calculated by the word2vec model to replace the keywords in the keyword vector, and use the TF-IDF feature selection function to calculate the weight of the word vector in the keyword vector, and get the replaced The keyword vector and the word vector weight corresponding to the word vector in the replaced keyword vector; according to the replaced keyword vector and word vector weight, the weighted matrix model of each document is obtained.

Common text feature selection functions in the prior art mainly include word contribution TC, information entropy/information gain, mutual information (Mutual Information) and χ ² statistics, wherein, information entropy/information gain only considers the contribution of features to the whole, It cannot be specific to each category, so information entropy/information gain is only suitable for selecting "global" features, not suitable for selecting "local" features, resulting in the inability to combine contextual information adjacent to words for calculation, which is not conducive to extracting from text Extract keywords; mutual information is only sensitive to critical feature words, so it often tends to affect the impact of rare words on text classification, and the words in the electrical work sheet are generally everyday words and not rare words, so mutual information is not suitable for calculating this Keywords in the embodiment; χ ² statistics can only obtain accurate results under the condition of χ ² distribution that the correlation between the feature words and categories of the text is satisfied, and when this assumption is not satisfied, the results obtained are consistent with the actual situation It is very different, and the words in the incoming electrician's order cannot guarantee to satisfy the χ ² distribution, so the χ ² distribution is not suitable for being used to extract keywords in this embodiment. Based on the above analysis, in this embodiment, the TF-IDF value in the word contribution TC is used as the feature selection function to calculate the weight of the word vector in the keyword vector. TF (Term Frequency) in the TF-IDF value is a measure of the local importance of a word, indicating the word frequency of the word in the document. The larger the TF value, the greater the contribution of the word to the document; IDF (Inverse Document Frequency) refers to Inverse text frequency index, the formula is log(D/D _w ), D represents the total number of texts, D _w is the number of texts in which the word w has appeared, and IDF is the cross-entropy (Kullback-Leibler Divergence) of the probability distribution of keywords under a specific condition . Therefore, the TF-IDF value measures the importance of words from the frequency of words in the text and the distribution in the corpus.

By using the word vector calculated by the word2vec model to replace the keywords in the keyword vector calculated by the TextRank model, and using the TF-IDF feature selection function to calculate the weight of the word vector, a text weighting matrix that can more accurately represent the text is obtained The model not only solves the problem of polysemy, but also combines the semantic information of the context to calculate the weight of the word vector from the frequency of the word in the document and the distribution in the corpus, which improves the accuracy of the word vector to represent the text. It achieves the effect of using the text weighted matrix model to represent documents, and further achieves the effect of improving the accuracy of the similarity between multiple documents.

Optionally, in the method for processing customer information provided in the embodiment of the present application, substituting multiple documents into the TextRank model for calculation, and obtaining the keyword set of each document includes: performing the first step on each of the multiple documents The first process is to obtain the words with preset parts of speech as the candidate keywords of each document, wherein the first process includes at least the following processes: sentence segmentation processing, word segmentation processing, filtering processing and tagging part of speech processing; according to each document and each document Candidate keywords of each document to obtain the second corpus; use the TextRank model to convert the candidate keywords of each document into a keyword directed graph, and calculate the weight of the candidate keywords in each document to obtain the weighted key of each document Word directed graph; from the weighted keyword directed graph, candidate keywords whose weight is higher than the preset weight are obtained, and the keyword set of each document is obtained.

Generally, keywords can reflect the key information or core idea of the entire text. Since there is no need to train the corpus in advance when using the TextRank model or LDA model based on the word graph model, the existing natural language parsing technology often uses these two algorithms to extract keywords in the text, but when the new text When adding the calculation, because the scalability of the LDA model is poor, if you continue to use the LDA model for calculation, you need to retrain the model, and the TextRank model is more concise when calculating, so in this embodiment, the TextRank model is used as A method for extracting keywords from multiple documents.

In this embodiment, in order to obtain the keyword sets corresponding to multiple documents, it is necessary to first perform sentence segmentation processing, word segmentation processing, filtering processing, and part-of-speech processing on multiple documents in the original corpus to obtain candidate keywords for each document ; form the second corpus by the candidate keywords corresponding to each document; then use the TextRank model to construct the candidate keyword graph G=(V, E), use the candidate keyword as the node E of the keyword graph, and adopt the co-occurrence relationship (co -occurrence) to construct the edge V between any two nodes, and iteratively calculate the weight of each node, until the error rate of any two nodes in the keyword graph is less than the preset threshold, then stop iteratively calculating the weight of each node, and get The weighted keyword directed graph corresponding to each document; according to the weight value of each node in the weighted keyword directed graph, the multiple nodes in the weighted keyword directed graph are sorted, and the node weights are obtained from multiple nodes C candidate keywords whose value is greater than the preset weight value are used as the keyword set corresponding to each document, wherein, if the c keywords in the keyword set of the document are adjacent phrases in the text of the call ticket, then Combine adjacent keywords to get multi-word keywords.

By using the TextRank model to construct a weighted keyword directed graph corresponding to each document in multiple documents, the keyword set corresponding to multiple documents is obtained, which improves the ability to use keywords to represent text information, and achieves a more accurate representation of documents The effect of meaning, achieves the effect of improving the accuracy of the similarity between multiple documents.

其中，d代表预设的阻尼系数，a表示在加权关键词有向图中的关键词a，Out(b)表示在加权关键词有向图中关键词a指向的所有关键词的集合，In(a)表示在加权关键词有向图中所有指向关键词a的关键词的集合，b表示指向关键词a的关键词b，c表示关键词a指向的关键词c，S(b)表示关键词b的词影响力，tfidf_a表示关键词a的词频和逆文档频率的乘积，tfidf_c表示关键词c的词频和逆文档频率的乘积，w_ba表示在加权关键词有向图中关键词a和关键词b的权值，w_bc表示在加权关键词有向图中关键词b和关键词c的权值；对多个文档中的每个文档进行第二处理，获取第三语料库以及第三语料库中的主题词，其中，第二处理至少包括以下处理：分词处理、过滤处理、提取初始主题词处理和统计词频处理；依据词影响力得分的计算公式，迭代计算第三语料库中的主题词的词影响力得分，当第N次迭代计算中与第N-1次迭代的差值小于预设阈值时停止迭代计算，从第N次迭代计算的计算结果中，获取词影响力得分高于预设词影响力得分的主题词，得到每个文档的主题词集合。Optionally, in the method for processing customer information provided in the embodiment of the present application, obtaining the keyword set of each of the multiple documents includes: configuring the calculation formula of the word influence score of the word vector according to the TextRank model, and the calculation formula is as follows Shown:

Among them, d represents the preset damping coefficient, a represents the keyword a in the weighted keyword directed graph, Out(b) represents the set of all keywords pointed to by keyword a in the weighted keyword directed graph, and In (a) represents the set of all keywords pointing to keyword a in the weighted keyword directed graph, b represents keyword b pointing to keyword a, c represents keyword c pointing to keyword a, S(b) represents The word influence of keyword b, tfidf _a represents the product of the word frequency of keyword a and the inverse document frequency, tfidf _c represents the product of the word frequency of keyword c and the inverse document frequency, w _ba represents the key word in the weighted keyword directed graph The weight of word a and keyword b, w _bc represents the weight of keyword b and keyword c in the weighted keyword directed graph; the second processing is performed on each document in the plurality of documents, and the third corpus is obtained And the subject words in the third corpus, wherein, the second processing includes at least the following processing: word segmentation processing, filtering processing, extracting initial subject words processing and statistical word frequency processing; according to the calculation formula of the word influence score, iteratively calculate the third corpus The word influence score of the subject word, when the difference between the Nth iterative calculation and the N-1th iteration is less than the preset threshold, the iterative calculation is stopped, and the word influence is obtained from the calculation result of the Nth iterative calculation The subject words whose scores are higher than the preset word influence score are obtained the set of subject words of each document.

Nowadays, the method of extracting topic models is commonly used to identify topic information in large-scale documents, but it is difficult for topic models to find suitable topic projection latitudes in the face of dynamically growing texts, resulting in the inability to accurately represent the topic words of documents. At the same time, topic models At present, Wordnet Semantic Dictionary and HowNet Semantic Dictionary are used, but Wordnet Semantic Dictionary and HowNet Semantic Dictionary are semantic dictionaries in the Chinese and English fields. When facing the highly professional text processing field, a large number of professional words will not be included in the dictionary in time As a result, the similarity of some words cannot be calculated and the problem of polysemy cannot be solved well. In the training process of the word vector, the context information of the word is captured. The context information of the word can not only bind the relationship between the word and the word, but also solve the semantic blank problem caused by the lack of professional vocabulary in the dictionary. Therefore, , in this embodiment, word vectors are used to measure the semantic relationship between words, so as to obtain a set of subject words corresponding to each document in multiple documents.

In this embodiment, because word vectors and keywords are introduced in the process of obtaining the subject word set corresponding to the document, when calculating the word influence between words, the original word influence formula Making improvements, the improved word influence score formula is as follows:

Among them, d represents the preset damping coefficient, a represents the keyword a in the weighted keyword directed graph, Out(b) represents the set of all keywords pointed to by keyword a in the weighted keyword directed graph, and In (a) represents the set of all keywords pointing to keyword a in the weighted keyword directed graph, b represents keyword b pointing to keyword a, c represents keyword c pointing to keyword a, S(b) represents The word influence of keyword b, tfidf _a represents the product of the word frequency of keyword a and the inverse document frequency, tfidf _c represents the product of the word frequency of keyword c and the inverse document frequency, w _ba represents the key word in the weighted keyword directed graph The weight of word a and keyword b, w _bc represents the weight of keyword b and keyword c in the weighted keyword directed graph. After determining the calculation formula of word influence, word segmentation processing, filtering processing, extraction of initial subject words and statistical word frequency processing are performed on multiple documents to obtain the third corpus and the subject words in the third corpus; finally, based on word influence The calculation formula of the power score, iteratively calculates the word influence score of each topic word in the third corpus until the difference between the Nth iteration calculation and the N-1th iteration is less than the preset threshold, or reaches the maximum iteration number of times, the iterative calculation is stopped, and from the calculation results of the Nth iterative calculation, the keyword whose word influence score is higher than the preset word influence score is obtained in each document, and the corresponding keyword set of each document is obtained.

By introducing the calculation formula of word vectors and keywords when calculating the word influence of keywords, using word vectors and keywords to measure the relationship between keywords, and improving the calculation formula for calculating the influence of words to calculate the weight of keywords , the set of keywords corresponding to each document is obtained, and the calculated keywords are combined with the contextual semantic information to achieve the effect of more accurately extracting the keywords in multiple documents, and further achieve the goal of improving the similarity between multiple documents effect on accuracy.

Optionally, in the method for processing customer information provided in the embodiment of the present application, according to the subject word set of each document and the weighted matrix model of each document, the relationship between each document and each other document among multiple documents is calculated The target similarity includes: constructing a bipartite graph model based on the keyword set of each document to calculate the first similarity between the keyword set of each document and the keyword set of each other document; according to the minimum The square distance formula calculates the second similarity between the weighted matrix model of each document and the weighted matrix model of each other document; calculates the relationship between each document and each other document based on the first similarity and the second similarity similarity between targets.

In order to calculate the target similarity between each document and each other document, it is necessary to first construct a bipartite graph model using the subject term set of each document, and calculate the subject term set of each document and each other document subject term The first degree of similarity between sets. The second similarity between the weighted matrix model of each document and the weighted matrix model of each other document is then calculated using the least squares distance formula of the matrix. For example, the specific steps to calculate the similarity between the text matrix doc ₁ (N×m dimensional matrix, N is the word vector dimension) and doc ₂ (N×n dimensional matrix) are: perform orthogonal triangular decomposition on the matrix doc ₁ , Obtain the orthogonal matrix X; calculate X ^T doc ₂ and perform orthogonal triangular decomposition to obtain the difference matrix D between matrix doc ₁ and matrix doc ₂ ; calculate the distance between matrix doc ₁ and matrix doc ₂ (ie, the second similarity) ,Calculated as follows:

Among them, a _ij is the element of row i and column j of difference matrix D. Finally, according to the first similarity and the second similarity, calculate the target similarity between each document and each other document, the calculation formula of the target similarity is as follows:

similarity(doc ₁ ,doc ₂ )=αDSM(doc ₁ ,doc ₂ )+βTSM(doc ₁ ,doc ₂ )

Among them, doc ₁ and doc ₂ are any two documents, DSM(doc ₁ ,doc ₂ ) represents the first similarity between matrix doc ₁ and matrix doc ₂ , TSM(doc ₁ ,doc ₂ ) represents matrix doc ₁ and The second similarity between the matrices doc ₂ , α and β are the weights of the first similarity and the second similarity respectively.

By combining the keyword set of each document and the weighted matrix model, the target similarity between each document and each other document is calculated, which not only avoids the influence of semantic gaps, but also combines the contextual semantic information in the document, so that The target similarity between each document and each other document is more accurate, and the effect of improving the accuracy of the similarity between multiple documents is achieved.

Optionally, in the method for processing customer information provided in the embodiment of the present application, a bipartite graph model is constructed according to the set of subject terms of each document, so as to calculate the difference between the set of subject terms of each document and the set of subject terms of each other document The first similarity between each document includes: constructing a bipartite graph model between each document and each other document based on the keyword set of each document; using the Hungarian algorithm to calculate the bipartite graph model between each document and each other document Matching maximum weight, the bipartite matching maximum weight is used as the first similarity between the subject word set of each document and the subject word set of each other document.

In this embodiment, in order to calculate the first similarity between the keyword set of each document and the keyword set of each other document, after obtaining the keyword set of each document, use the keyword set of each document The collection builds a bipartite graph model between each document and every other document, that is, uses the bipartite graph model between two documents to represent the correlation between two documents. The specific process of building a bipartite graph model between two documents is as follows: Given two topic sets T_d1 and T_d2 of two documents, construct a bipartite graph B(T_d1, T_d2), |V(T_d1)| represents the topic set T_d1 , |V(T_d2)| represents the node of the topic set T_d2, use b(u) to represent the relevant node u in B(T_d1, T_d2); for each node u∈ in the bipartite graph B(T_d1, T_d2) V(T_d1), in the node v∈V(T_d2), select the node v with the greatest similarity with u to form an edge. If there are multiple nodes v, the similarity between node u and node v is w(b(u) , b(v)) is the largest, then multiple edges are formed; for each node v∈V(T_d2), select the node with the highest similarity with v in the node u∈V(T_d1) to form a connection, if the connection If it already exists, there is no need to add a connection again; then use the Hungarian algorithm to find the edge with the highest similarity among the connected edges, that is, start from an unmatched point, and keep looking for the augmented path. If the augmented path is found, then Match the matching edges and matching points on the augmented path. If the augmented path cannot be found, confirm that the bipartite graph has reached the maximum match; after traversing all the nodes in the bipartite graph, obtain unmatched nodes , build a new bipartite graph and perform matching again until there are no unmatched nodes in the bipartite graph, then obtain the maximum similarity w(b(u),b(v)) from the matched bipartite graph (ie maximum value). Finally, the maximum weight after the bipartite graph matching is used as the first similarity between the subject term set of each document and the subject term set of each other document.

By introducing the bipartite graph model and the Hungarian algorithm, the first similarity between the subject word sets of each document in multiple documents is calculated, so that the first similarity between the subject word sets of each document in multiple documents is more accurate. Accurate, achieving the effect of improving the accuracy of the similarity between multiple documents.

Optionally, in the method for processing customer information provided in the embodiment of the present application, obtaining the target information of multiple customers from the clustered document collection includes: according to the method of obtaining the subject word collection, for each clustered The document collection is processed to obtain the subject word collection of each clustered document collection; according to the method of obtaining the keyword collection, each clustered document collection is processed to obtain the key word of each clustered document collection A word set; according to the subject word set of each clustered document set and the keyword set of each clustered document set, target information of multiple customers is determined.

In order to mine customers’ potential value information (i.e. target information) from multiple documents, it is first necessary to extract the subject terms from each cluster document in multiple cluster documents according to the method of obtaining the subject term set from the document. The set of subject words corresponding to each cluster document is used as the customer information of the customer call information business subclass. Then, according to the method of obtaining keyword sets from documents, keywords are extracted from each cluster document in multiple cluster documents, and the keyword set corresponding to each cluster document is used as the customer of the customer call information question summary class information. Finally, combined with the customer information of the customer call information business subclass and the customer information of the customer call information problem summary class, the target information extracted from multiple call work orders of multiple customers is obtained.

By obtaining the subject word set and the keyword set from the document, the clustered document is processed to obtain the customer information of the customer call information business subclass and the customer information of the customer call information problem summary class (ie target information), It realizes the full consideration of the context information in the incoming work order when extracting the target information, achieves the effect of mining the target information that customers pay more attention to from the incoming work order, and further achieves the effect of extracting more accurate customer potential value information from the incoming work order .

Optionally, in the method for processing customer information provided in the embodiment of the present application, according to the keyword set of each document and the weighted matrix model of each document, calculate the difference between each document and each other document in multiple documents After the target similarity among multiple documents, the above method also includes: adding the target similarity between each document and every other document in the multiple documents to obtain the third similarity of each document; from the multiple documents, Acquiring target documents with a third similarity higher than the preset similarity to obtain a target document set; pushing the target document set to the target object.

After calculating the target similarity between each of the multiple documents, the solution implemented in this embodiment also includes obtaining at least one of the most popular incoming work orders from the multiple incoming work orders, that is, the core content of the work order is in At least one incoming work order that appears most frequently among multiple incoming work orders: first calculate the sum of the target similarities between each document and each other document in multiple documents, and obtain the third similarity corresponding to each document ; From multiple documents, obtain the target document whose third similarity is higher than the preset similarity, and obtain the target document set, which is at least one of the most popular work orders among the multiple call work orders; finally, the target document The collection is pushed to the relevant staff (that is, the target audience) to improve the quality of customer service and provide customers with more precise marketing to meet the urgent needs of customers.

By calculating the third similarity of multiple incoming call work orders, the incoming call work orders that customers pay more attention to are obtained, and the incoming call work orders that customers pay more attention to are pushed to relevant staff, which is beneficial to improve the service quality of staff to customers, and It is conducive to providing customers with precise marketing to meet the urgent needs of customers, achieving the effect of more accurately extracting the incoming call work orders that customers pay more attention to, and further achieving the effect of improving customer satisfaction.

Optionally, in this embodiment, the process of calculating the similarity between multiple weighted matrix models in this solution can be shown in Figure 2. First, preprocessing and word segmentation processing are performed on multiple documents in the corpus, and the obtained The first corpus, then use the TextRank model to extract the keyword set corresponding to each document in multiple documents from the first corpus, use the Word2vec model to train the first corpus to obtain word vectors, and then use the keyword set corresponding to each document Each keyword in is replaced with the corresponding word vector to obtain the keyword vector, and the weight of the keyword vector is calculated based on the TF-IDF value, and the text matrix model of each document is obtained, using the minimum matrix square distance multiple The similarity between text matrix models, and finally, the similarity between multiple text matrix models is substituted into the calculation formula for calculating text similarity, and the text similarity value between multiple documents is calculated and obtained.

Optionally, in this embodiment, the main workflow of mining value information from multiple call orders of multiple customers in this solution may be as shown in FIG. 3 . In the first step, the word vector model based on Word2vec performs semantic retrieval on the electrical work orders after preprocessing; in the second step, based on the TextRank model, the keywords in multiple incoming work orders are extracted, and the influence of words based on keywords is mined for multiple The subject words in the call work order; the third step, based on the text weighted matrix model to represent the text content of multiple call work orders, and the subject words based on the text graph model to represent the text content of multiple call work orders; the fourth step, based on The matrix least squares distance algorithm calculates the keyword model similarity between multiple text weighted matrices, and the bipartite graph matching algorithm based on Hungarian calculates the topic model similarity between multiple topic word sets; the fifth step, based on aggregation The class analysis algorithm mines potential intentions of customers, and mines popular work orders among multiple incoming work orders based on the calculated text similarity.

Optionally, in this embodiment, the workflow of mining value information from multiple call orders of multiple customers in this solution may be as shown in FIG. 4 . The first step is to perform text preprocessing on the original corpus, and use the word vector model based on Word2vec to calculate the processed corpus to obtain the word vector corresponding to the corpus. The second step is to use the TextRank model to extract the keywords in the processed corpus to obtain a text keyword set, and iteratively calculate the weighted score of keywords according to the keyword frequency/inverse document word frequency to obtain a text topic set. In the third step, the weighted text matrix is constructed according to the text keyword set, and the bipartite graph model is constructed and the topic similarity is calculated according to the text topic set. The fourth step is to use the matrix least squares distance algorithm to calculate the text semantic similarity between text weighted matrices, use the text topic similarity matching to calculate the text topic similarity, and then calculate the text semantic similarity and text topic similarity according to the text semantic similarity and text topic similarity. similarity. The fifth step is to preprocess multiple call tickets, use the PAM clustering algorithm to analyze and calculate the preprocessed multiple call tickets, extract subject words and keywords from multiple call tickets, and finally, according to Extract subject words and keywords from multiple incoming work orders to mine customer call intentions, and mine Top work orders (that is, popular work orders) among multiple incoming work orders according to the text similarity in the fourth step.

It should be noted that the steps shown in the flowcharts of the accompanying drawings may be performed in a computer system, such as a set of computer-executable instructions, and that although a logical order is shown in the flowcharts, in some cases, The steps shown or described may be performed in an order different than here.

The embodiment of the present application also provides a client information processing device. It should be noted that the client information processing device in the embodiment of the present application can be used to execute the method for processing client information provided in the embodiment of the present application. The following introduces the client information processing device provided by the embodiment of the present application.

Fig. 5 is a schematic diagram of an apparatus for processing customer information according to an embodiment of the present application. As shown in FIG. 5 , the apparatus includes: a first acquisition unit 501 , a second acquisition unit 502 , a third acquisition unit 503 , a calculation unit 504 , a fourth acquisition unit 505 and a fifth acquisition unit 506 .

Specifically, the first acquisition unit 501 is configured to encode customer information in multiple incoming work orders of multiple customers to obtain multiple documents corresponding to multiple incoming work orders, wherein the encoding formats of the multiple documents are the same .

The second obtaining unit 502 is configured to obtain a weighted matrix model of each document in the plurality of documents.

The third obtaining unit 503 is configured to obtain a subject word set of each document in the plurality of documents.

The calculation unit 504 is configured to calculate the target similarity between each document and each other document among the plurality of documents according to the keyword set of each document and the weighted matrix model of each document.

The fourth acquiring unit 505 is configured to cluster multiple documents based on the PAM clustering algorithm according to the target similarity to obtain a clustered document set.

The fifth obtaining unit 506 is configured to obtain target information of multiple customers from the clustered document collection.

The customer information processing device provided in the embodiment of the present application is used to encode the customer information in the multiple call work orders of multiple customers through the first acquisition unit 501, and obtain multiple documents corresponding to the multiple call work orders , wherein the encoding formats of multiple documents are the same; the second acquisition unit 502 is used to acquire the weighted matrix model of each document in the multiple documents; the third acquisition unit 503 is used to acquire the subject of each document in the multiple documents word set; calculation unit 504, for calculating the target similarity between each document and each other document in multiple documents according to the subject word set of each document and the weighted matrix model of each document; the fourth acquisition unit 505, for clustering multiple documents based on the PAM clustering algorithm according to the target similarity, to obtain a clustered document set; the fifth acquisition unit 506, for acquiring multiple customers from the clustered document set The target information solves the problem that manual labeling or natural language processing technology is used to extract the value information of the incoming work order, resulting in inaccurate information extracted from the incoming work order. By calculating the weighted matrix model of each document and the subject word set of each document in multiple incoming call tickets, to calculate the target similarity between multiple documents, so as to extract target information in multiple documents, the algorithm is implemented Automatically extract target information from multiple documents, avoiding the impact of manual labeling of multiple documents on the information quality of target information, and achieve the effect of extracting more accurate target information from multiple call work orders.

Optionally, in the client information processing device provided in the embodiment of the present application, the above-mentioned second acquisition unit 502 includes: a first processing subunit, configured to delete preset character strings in multiple documents, and Carry out word segmentation and filtering processing for each document to obtain the first corpus; the first calculation subunit is used to substitute the first corpus into the word2vec model for calculation to obtain the word vector of each document; the second calculation subunit is used for Substituting multiple documents into the TextRank model for calculation to obtain the keyword set of each document; the first acquisition subunit is used to use The greedy selection algorithm sorts the keywords in the keyword set of each document to obtain the keyword vector of each document; the third calculation subunit is used to replace the keywords in the keyword vector with the word vector, and use TF -IDF feature selection function calculates the weight of the word vector in the keyword vector, and obtains the word vector weight corresponding to the word vector in the replaced keyword vector and the replaced keyword vector; the second acquisition subunit is used for Obtain the weighted matrix model of each document according to the replaced keyword vector and word vector weights.

Optionally, in the client information processing device provided in the embodiment of the present application, the above-mentioned second calculation subunit includes: a processing module, configured to perform the first processing on each of the multiple documents, and obtain the preset part-of-speech Words are used as candidate keywords for each document, wherein the first processing includes at least the following processing: sentence segmentation processing, word segmentation processing, filtering processing, and tagging part-of-speech processing; the first acquisition module is used for each document and each document according to The candidate keywords are obtained from the second corpus; the first calculation module is used to convert the candidate keywords of each document into a keyword directed graph using the TextRank model, and calculate the weight of the candidate keywords in each document to obtain The weighted keyword directed graph of each document; the second obtaining module is used to obtain candidate keywords whose weight value is higher than a preset weight value from the weighted keyword directed graph, and obtain a keyword set of each document.

其中，d代表预设的阻尼系数，a表示在加权关键词有向图中的关键词a，Out(b)表示在加权关键词有向图中关键词a指向的所有关键词的集合，In(a)表示在加权关键词有向图中所有指向关键词a的关键词的集合，b表示指向关键词a的关键词b，c表示关键词a指向的关键词c，S(b)表示关键词b的词影响力，tfidf_a表示关键词a的词频和逆文档频率的乘积，tfidf_c表示关键词c的词频和逆文档频率的乘积，w_ba表示在加权关键词有向图中关键词a和关键词b的权值，w_bc表示在加权关键词有向图中关键词b和关键词c的权值；第二处理子单元，用于对多个文档中的每个文档进行第二处理，获取第三语料库以及第三语料库中的主题词，其中，第二处理至少包括以下处理：分词处理、过滤处理、提取初始主题词处理和统计词频处理；第四计算子单元，用于依据词影响力得分的计算公式，迭代计算第三语料库中的主题词的词影响力得分，当第N次迭代计算中与第N-1次迭代的差值小于预设阈值时停止迭代计算，从第N次迭代计算的计算结果中，获取词影响力得分高于预设词影响力得分的主题词，得到每个文档的主题词集合。Optionally, in the client information processing device provided in the embodiment of the present application, the above-mentioned third acquisition unit 503 includes: a configuration subunit for configuring the calculation formula of the word influence score of the word vector according to the TextRank model, and the calculation formula is as follows Shown:

Among them, d represents the preset damping coefficient, a represents the keyword a in the weighted keyword directed graph, Out(b) represents the set of all keywords pointed to by keyword a in the weighted keyword directed graph, and In (a) represents the set of all keywords pointing to keyword a in the weighted keyword directed graph, b represents keyword b pointing to keyword a, c represents keyword c pointing to keyword a, S(b) represents The word influence of keyword b, tfidf _a represents the product of the word frequency of keyword a and the inverse document frequency, tfidf _c represents the product of the word frequency of keyword c and the inverse document frequency, w _ba represents the key word in the weighted keyword directed graph The weight value of word a and keyword b, w _bc represents the weight value of keyword b and keyword c in the weighted keyword directed graph; the second processing subunit is used for each document in a plurality of documents The second process is to obtain the third corpus and the subject words in the third corpus, wherein the second process at least includes the following processing: word segmentation processing, filtering processing, extracting initial subject words processing and statistical word frequency processing; the fourth calculation subunit, using Based on the calculation formula of the word influence score, iteratively calculate the word influence score of the subject words in the third corpus, and stop the iterative calculation when the difference between the Nth iterative calculation and the N-1th iteration is less than the preset threshold , from the calculation results of the Nth iterative calculation, obtain the keyword whose word influence score is higher than the preset word influence score, and obtain the keyword set of each document.

Optionally, in the client information processing device provided in the embodiment of the present application, the calculation unit 504 includes: a construction subunit, configured to construct a bipartite graph model according to the subject word set of each document, so as to calculate the The first similarity between the subject word set and the subject term set of each other document; the fifth calculation subunit is used to calculate the weighted matrix model of each document and each other document according to the least squares distance formula of the matrix The second similarity between the weighted matrix models; the sixth calculation subunit is used to calculate the target similarity between each document and each other document according to the first similarity and the second similarity.

Optionally, in the client information processing device provided in the embodiment of the present application, the above-mentioned first construction subunit includes: a construction module, configured to construct each document and each other document according to the subject word set of each document Bipartite graph model; the second calculation module is used to calculate the maximum weight of bipartite matching between each document and each other document using the Hungarian algorithm, and use the maximum weight of bipartite matching as the keyword set of each document and The first degree of similarity between the set of subject terms of each of the other documents.

Optionally, in the client information processing device provided in the embodiment of the present application, the above-mentioned fifth acquisition unit 506 includes: a third processing subunit, configured to, according to the method of acquiring a set of subject words, classify each clustered document The set is processed to obtain the keyword set of each clustered document set; the fourth processing subunit is used to process each clustered document set according to the method of obtaining the keyword set, and obtain each clustered The keyword set of the document collection after the classification; determine the subunit, which is used to determine the target information of multiple customers according to the keyword collection of each clustered document collection and the keyword collection of each clustered document collection .

Optionally, in the client information processing device provided in the embodiment of the present application, the above-mentioned device further includes: a sixth acquisition unit, which is used to calculate the relationship between each document among the multiple documents and other documents according to the set of subject terms and the weighted matrix model. After the target similarity between each document, add the target similarity between each document and each other document in multiple documents to obtain the third similarity of each document; the seventh acquisition unit is used to From the plurality of documents, a target document with a third similarity higher than a preset similarity is obtained to obtain a target document set; a pushing unit is configured to push the target document set to a target object.

The customer information processing device includes a processor and a memory, and the above-mentioned first acquisition unit 501, second acquisition unit 502, third acquisition unit 503, calculation unit 504, fourth acquisition unit 505, and fifth acquisition unit 506 all serve as The program unit is stored in the memory, and the processor executes the program unit stored in the memory to realize corresponding functions.

The processor includes a kernel, and the kernel fetches corresponding program units from the memory. One or more kernels can be set, and the accuracy of the similarity between multiple documents can be improved by adjusting kernel parameters.

Memory may include non-permanent memory in computer-readable media, in the form of random access memory (RAM) and/or non-volatile memory, such as read-only memory (ROM) or flash memory (flash RAM), memory including at least one memory chip.

An embodiment of the present invention provides a computer-readable storage medium on which a program is stored, and when the program is executed by a processor, a method for processing customer information is realized.

An embodiment of the present invention provides a processor, and the processor is used to run a program, wherein the method for processing customer information is executed when the program is running.

As shown in Figure 6, an embodiment of the present invention provides an electronic device, which includes a processor, a memory, and a program stored on the memory and operable on the processor. When the processor executes the program, the following steps are implemented: Coding the customer information in the customer's multiple incoming work orders to obtain multiple documents corresponding to the multiple incoming work orders, wherein the encoding format of the multiple documents is the same; obtain the weighted matrix model of each document in the multiple documents; Acquiring the keyword set of each document in the plurality of documents; calculating the target similarity between each document in the plurality of documents and each other document according to the keyword set of each document and the weighted matrix model of each document; According to the target similarity, multiple documents are clustered based on the PAM clustering algorithm to obtain a clustered document set; the target information of multiple customers is obtained from the clustered document set.

When the processor executes the program, the following steps are also implemented: obtaining the weighted matrix model of each of the multiple documents includes: deleting the preset character strings in the multiple documents, and performing word segmentation and filtering for each of the multiple documents, Obtain the first corpus; Substituting the first corpus into the word2vec model for calculation to obtain the word vector of each document; Substituting multiple documents into the TextRank model for calculation to obtain the keyword set of each document; according to the key words of each document The similarity between the word set and the keyword set of each other document, use the greedy selection algorithm to sort the keywords in the keyword set of each document, and obtain the keyword vector of each document; use the word vector to replace the keyword vector keywords in , and use the TF-IDF feature selection function to calculate the weight of the word vector in the keyword vector, and obtain the replaced keyword vector and the word vector weight corresponding to the word vector in the replaced keyword vector; Obtain the weighted matrix model of each document according to the replaced keyword vector and word vector weights.

When the processor executes the program, the following steps are also implemented: Substituting multiple documents into the TextRank model for calculation, and obtaining the keyword set of each document includes: first processing each document in the multiple documents, and obtaining the preset part-of-speech Words are used as candidate keywords for each document, wherein the first processing includes at least the following processing: sentence segmentation processing, word segmentation processing, filtering processing, and part-of-speech tagging processing; according to each document and the candidate keywords of each document, obtain the second Corpus; use the TextRank model to convert the candidate keywords of each document into a keyword directed graph, and calculate the weight of the candidate keywords in each document to obtain the weighted keyword directed graph of each document; from the weighted key In the word directed graph, candidate keywords whose weights are higher than the preset weights are obtained, and the keyword set of each document is obtained.

其中，d代表预设的阻尼系数，a表示在加权关键词有向图中的关键词a，Out(b)表示在加权关键词有向图中关键词a指向的所有关键词的集合，In(a)表示在加权关键词有向图中所有指向关键词a的关键词的集合，b表示指向关键词a的关键词b，c表示关键词a指向的关键词c，S(b)表示关键词b的词影响力，tfidf_a表示关键词a的词频和逆文档频率的乘积，tfidf_c表示关键词c的词频和逆文档频率的乘积，w_ba表示在加权关键词有向图中关键词a和关键词b的权值，w_bc表示在加权关键词有向图中关键词b和关键词c的权值；对多个文档中的每个文档进行第二处理，获取第三语料库以及第三语料库中的主题词，其中，第二处理至少包括以下处理：分词处理、过滤处理、提取初始主题词处理和统计词频处理；依据词影响力得分的计算公式，迭代计算第三语料库中的主题词的词影响力得分，当第N次迭代计算中与第N-1次迭代的差值小于预设阈值时停止迭代计算，从第N次迭代计算的计算结果中，获取词影响力得分高于预设词影响力得分的主题词，得到每个文档的主题词集合。When the processor executes the program, the following steps are also implemented: obtaining the subject word set of each document in the plurality of documents includes: configuring the calculation formula of the word influence score of the word vector according to the TextRank model, and the calculation formula is as follows:

Among them, d represents the preset damping coefficient, a represents the keyword a in the weighted keyword directed graph, Out(b) represents the set of all keywords pointed to by keyword a in the weighted keyword directed graph, and In (a) represents the set of all keywords pointing to keyword a in the weighted keyword directed graph, b represents keyword b pointing to keyword a, c represents keyword c pointing to keyword a, S(b) represents The word influence of keyword b, tfidf _a represents the product of the word frequency of keyword a and the inverse document frequency, tfidf _c represents the product of the word frequency of keyword c and the inverse document frequency, w _ba represents the key word in the weighted keyword directed graph The weight of word a and keyword b, w _bc represents the weight of keyword b and keyword c in the weighted keyword directed graph; the second processing is performed on each document in the plurality of documents, and the third corpus is obtained And the subject words in the third corpus, wherein, the second processing includes at least the following processing: word segmentation processing, filtering processing, extracting initial subject words processing and statistical word frequency processing; according to the calculation formula of the word influence score, iteratively calculate the third corpus The word influence score of the subject word, when the difference between the Nth iterative calculation and the N-1th iteration is less than the preset threshold, the iterative calculation is stopped, and the word influence is obtained from the calculation result of the Nth iterative calculation The subject words whose scores are higher than the preset word influence score are obtained the set of subject words of each document.

When the processor executes the program, the following steps are also implemented: according to the subject word set of each document and the weighted matrix model of each document, calculating the target similarity between each document in the multiple documents and each other document includes: according to each A bipartite graph model is constructed from the subject term sets of documents to calculate the first similarity between the subject term sets of each document and the subject term sets of each other document; according to the least square distance formula of the matrix, calculate each The second similarity between the weighted matrix model of the document and the weighted matrix model of each other document; according to the first similarity and the second similarity, calculate the target similarity between each document and each other document.

When the processor executes the program, the following steps are also implemented: constructing a bipartite graph model according to the subject term set of each document, to calculate the first similarity between the subject term set of each document and the subject term set of each other document, including : According to the keyword set of each document, construct a bipartite graph model of each document and each other document; use the Hungarian algorithm to calculate the bipartite matching maximum weight between each document and each other document, and divide the bipartite The maximum matching weight is used as the first similarity between the subject word set of each document and the subject word set of each other document.

When the processor executes the program, the following steps are also implemented: obtaining the target information of multiple customers from the clustered document collection includes: processing each clustered document collection according to the method of obtaining the subject word collection, and obtaining each The keyword set of the clustered document set; according to the method of obtaining the keyword set, process each clustered document set to obtain the keyword set of each clustered document set; according to each cluster The subject word set of the final document set and the keyword set of each clustered document set are used to determine the target information of multiple customers.

The following steps are also implemented when the processor executes the program: after calculating the target similarity between each document and each other document in the plurality of documents according to the subject word set of each document and the weighted matrix model of each document, the above-mentioned The method also includes: adding the target similarity between each document and each other document in the plurality of documents to obtain a third similarity of each document; from the plurality of documents, obtaining the third similarity higher than the preset Set the target document of similarity to obtain the target document set; push the target document set to the target object.

The devices in this article can be servers, PCs, PADs, mobile phones, etc.

The present application also provides a computer program product, which, when executed on a data processing device, is suitable for executing a program that is initialized with the following method steps: encoding and processing customer information in multiple incoming call orders of multiple customers to obtain Multiple documents corresponding to multiple incoming work orders, wherein the encoding formats of the multiple documents are the same; obtain the weighted matrix model of each document in the multiple documents; obtain the subject word set of each document in the multiple documents; according to each The keyword set of the document and the weighted matrix model of each document calculate the target similarity between each document in multiple documents and each other document; according to the target similarity, multiple documents are clustered based on the PAM clustering algorithm class to obtain a clustered document collection; and obtain target information of multiple customers from the clustered document collection.

When executed on a data processing device, it is also suitable for executing a program initialized with the following method steps: obtaining a weighted matrix model of each document in a plurality of documents includes: deleting preset character strings in a plurality of documents, and Each document in the document is segmented and filtered to obtain the first corpus; the first corpus is substituted into the word2vec model for calculation to obtain the word vector of each document; multiple documents are substituted into the TextRank model for calculation to obtain each document The keyword set of each document; according to the similarity between the keyword set of each document and the keyword set of each other document, use the greedy selection algorithm to sort the keywords in the keyword set of each document, and obtain the keywords of each document Keyword vector; use the word vector to replace the keywords in the keyword vector, and use the TF-IDF feature selection function to calculate the weight of the word vector in the keyword vector to obtain the replaced keyword vector and the replaced keyword vector The word vector weights corresponding to the word vectors in ; obtain the weighted matrix model of each document according to the replaced keyword vectors and word vector weights.

When executing on a data processing device, it is also suitable for executing a program that is initialized with the following method steps: Substituting multiple documents into the TextRank model for calculation, obtaining the keyword set of each document includes: for each of the multiple documents The first processing is performed on the document, and the words of the preset part of speech are obtained as the candidate keywords of each document, wherein the first processing includes at least the following processing: sentence segmentation processing, word segmentation processing, filtering processing and tagging part of speech processing; according to each document and The candidate keywords of each document are obtained from the second corpus; the candidate keywords of each document are converted into a keyword directed graph using the TextRank model, and the weights of the candidate keywords in each document are calculated to obtain each document The weighted keyword directed graph; from the weighted keyword directed graph, obtain the candidate keywords whose weight value is higher than the preset weight value, and obtain the keyword set of each document.

其中，d代表预设的阻尼系数，a表示在加权关键词有向图中的关键词a，Out(b)表示在加权关键词有向图中关键词a指向的所有关键词的集合，In(a)表示在加权关键词有向图中所有指向关键词a的关键词的集合，b表示指向关键词a的关键词b，c表示关键词a指向的关键词c，S(b)表示关键词b的词影响力，tfidf_a表示关键词a的词频和逆文档频率的乘积，tfidf_c表示关键词c的词频和逆文档频率的乘积，w_ba表示在加权关键词有向图中关键词a和关键词b的权值，w_bc表示在加权关键词有向图中关键词b和关键词c的权值；对多个文档中的每个文档进行第二处理，获取第三语料库以及第三语料库中的主题词，其中，第二处理至少包括以下处理：分词处理、过滤处理、提取初始主题词处理和统计词频处理；依据词影响力得分的计算公式，迭代计算第三语料库中的主题词的词影响力得分，当第N次迭代计算中与第N-1次迭代的差值小于预设阈值时停止迭代计算，从第N次迭代计算的计算结果中，获取词影响力得分高于预设词影响力得分的主题词，得到每个文档的主题词集合。When executing on the data processing device, it is also suitable for executing a program that is initialized with the following method steps: obtaining the set of subject words of each document in a plurality of documents includes: configuring the calculation formula of the word influence score of the word vector according to the TextRank model, The calculation formula is as follows:

Among them, d represents the preset damping coefficient, a represents the keyword a in the weighted keyword directed graph, Out(b) represents the set of all keywords pointed to by keyword a in the weighted keyword directed graph, and In (a) represents the set of all keywords pointing to keyword a in the weighted keyword directed graph, b represents keyword b pointing to keyword a, c represents keyword c pointing to keyword a, S(b) represents The word influence of keyword b, tfidf _a represents the product of the word frequency of keyword a and the inverse document frequency, tfidf _c represents the product of the word frequency of keyword c and the inverse document frequency, w _ba represents the key word in the weighted keyword directed graph The weight of word a and keyword b, w _bc represents the weight of keyword b and keyword c in the weighted keyword directed graph; the second processing is performed on each document in the plurality of documents, and the third corpus is obtained And the subject words in the third corpus, wherein, the second processing includes at least the following processing: word segmentation processing, filtering processing, extracting initial subject words processing and statistical word frequency processing; according to the calculation formula of the word influence score, iteratively calculate the third corpus The word influence score of the subject word, when the difference between the Nth iterative calculation and the N-1th iteration is less than the preset threshold, the iterative calculation is stopped, and the word influence is obtained from the calculation result of the Nth iterative calculation The subject words whose scores are higher than the preset word influence score are obtained the set of subject words of each document.

When executed on a data processing device, it is also suitable for executing a program that is initialized with the following method steps: according to the subject word set of each document and the weighted matrix model of each document, calculate the ratio of each document in multiple documents to each other The target similarity between documents includes: constructing a bipartite graph model based on the subject term set of each document to calculate the first similarity between the subject term set of each document and the subject term set of each other document; The least squares distance formula of the matrix calculates the second similarity between the weighted matrix model of each document and the weighted matrix model of each other document; according to the first similarity and second similarity, calculates the relationship between each document and other Target similarity between each document.

When executed on the data processing device, it is also suitable for executing a program initialized with the following method steps: constructing a bipartite graph model according to the subject term set of each document, to calculate the difference between the subject term set of each document and the subject term set of each other document The first similarity between the subject word sets includes: constructing a bipartite graph model between each document and each other document based on the subject term set of each document; using the Hungarian algorithm to calculate the difference between each document and each other document The maximum weight of bipartite matching is the first similarity between the subject word set of each document and the subject word set of each other document.

When executed on a data processing device, it is also suitable for executing a program that is initialized with the following method steps: obtaining the target information of multiple customers from the clustered document collection includes: according to the method for obtaining the subject word collection, for each clustered Process the document collection after the class, and obtain the subject word collection of each clustered document collection; process each clustered document collection according to the method of obtaining the keyword collection, and obtain each clustered document A keyword set of the set; according to the subject word set of each clustered document set and the keyword set of each clustered document set, target information of multiple customers is determined.

When executed on a data processing device, it is also suitable for executing a program that is initialized with the following method steps: according to the set of subject terms of each document and the weighted matrix model of each document, calculate the relationship between each document in multiple documents After the target similarity between documents, the above method also includes: adding the target similarity between each document and every other document in multiple documents to obtain the third similarity of each document; In the document, the target document whose third similarity is higher than the preset similarity is obtained to obtain a target document set; and the target document set is pushed to the target object.

Those skilled in the art should understand that the embodiments of the present application may be provided as methods, systems, or computer program products. Accordingly, the present application may take the form of an entirely hardware embodiment, an entirely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the present application may take the form of a computer program product embodied on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) having computer-usable program code embodied therein.

The present application is described with reference to flowcharts and/or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of the present application. It should be understood that each procedure and/or block in the flowchart and/or block diagram, and a combination of procedures and/or blocks in the flowchart and/or block diagram can be realized by computer program instructions. These computer program instructions may be provided to a general purpose computer, special purpose computer, embedded processor, or processor of other programmable data processing equipment to produce a machine such that the instructions executed by the processor of the computer or other programmable data processing equipment produce a An apparatus for realizing the functions specified in one or more procedures of the flowchart and/or one or more blocks of the block diagram.

These computer program instructions may also be stored in a computer-readable memory capable of directing a computer or other programmable data processing apparatus to operate in a specific manner, such that the instructions stored in the computer-readable memory produce an article of manufacture comprising instruction means, the instructions The device realizes the function specified in one or more procedures of the flowchart and/or one or more blocks of the block diagram.

These computer program instructions can also be loaded onto a computer or other programmable data processing device, causing a series of operational steps to be performed on the computer or other programmable device to produce a computer-implemented process, thereby The instructions provide steps for implementing the functions specified in the flow chart or blocks of the flowchart and/or the block or blocks of the block diagrams.

In a typical configuration, a computing device includes one or more processors (CPUs), input/output interfaces, network interfaces, and memory.

Memory may include non-permanent storage in computer readable media, in the form of random access memory (RAM) and/or nonvolatile memory such as read only memory (ROM) or flash RAM. The memory is an example of a computer readable medium.

Computer-readable media, including both permanent and non-permanent, removable and non-removable media, can be implemented by any method or technology for storage of information. Information may be computer readable instructions, data structures, modules of a program, or other data. Examples of computer storage media include, but are not limited to, phase change memory (PRAM), static random access memory (SRAM), dynamic random access memory (DRAM), other types of random access memory (RAM), read only memory (ROM), Electrically Erasable Programmable Read-Only Memory (EEPROM), Flash memory or other memory technology, Compact Disc Read-Only Memory (CD-ROM), Digital Versatile Disc (DVD) or other optical storage, Magnetic tape cartridge, tape disk storage or other magnetic storage device or any other non-transmission medium that can be used to store information that can be accessed by a computing device. As defined herein, computer-readable media excludes transitory computer-readable media, such as modulated data signals and carrier waves.

It should also be noted that the term "comprises", "comprises" or any other variation thereof is intended to cover a non-exclusive inclusion such that a process, method, article, or apparatus comprising a set of elements includes not only those elements, but also includes Other elements not expressly listed, or elements inherent in the process, method, commodity, or apparatus are also included. Without further limitations, an element defined by the phrase "comprising a ..." does not preclude the presence of additional identical elements in the process, method, article, or apparatus that includes the element.

Those skilled in the art should understand that the embodiments of the present application may be provided as methods, systems or computer program products. Accordingly, the present application can take the form of an entirely hardware embodiment, an entirely software embodiment or an embodiment combining software and hardware aspects. Furthermore, the present application may take the form of a computer program product embodied on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) having computer-usable program code embodied therein.

The above are only examples of the present application, and are not intended to limit the present application. For those skilled in the art, various modifications and changes may occur in this application. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present application shall be included within the scope of the claims of the present application.

Claims (10)

1. A method for processing customer information, comprising: Encoding the customer information in multiple incoming work orders of multiple customers to obtain multiple documents corresponding to the multiple incoming electric work orders, wherein the encoding formats of the multiple documents are the same; obtaining a weighted matrix model for each of the plurality of documents; Obtain a set of subject terms of each document in the plurality of documents; calculating the target similarity between each of the plurality of documents and each of the other documents according to the keyword set of each document and the weighted matrix model of each document; According to the target similarity, cluster the multiple documents based on the PAM clustering algorithm to obtain a clustered document set; The target information of multiple customers is obtained from the clustered document collection. 2. The method according to claim 1, wherein obtaining the weighted matrix model of each document in the plurality of documents comprises: Deleting the preset character strings in the plurality of documents, and performing word segmentation and filtering on each of the plurality of documents to obtain the first corpus; Substituting the first corpus into the word2vec model for calculation to obtain the word vector of each document; Substituting the multiple documents into the TextRank model for calculation, to obtain the keyword set of each document; According to the similarity between the keyword set of each document and the keyword set of each other document, use the greedy selection algorithm to sort the keywords in the keyword set of each document, and obtain the keyword vector of each document; Use the word vector to replace the keywords in the keyword vector, and use the TF-IDF feature selection function to calculate the weight of the word vector in the keyword vector, obtain the replaced keyword vector and the replaced The word vector weight corresponding to the word vector in the keyword vector; A weighted matrix model of each document is obtained according to the replaced keyword vector and the word vector weight. 3. The method according to claim 2, characterized in that, the multiple documents are substituted into the TextRank model to calculate, and the keyword set obtained for each document comprises: Perform the first processing on each document in the plurality of documents, and obtain the words with preset parts of speech as candidate keywords of each document, wherein the first processing includes at least the following processing: sentence segmentation processing, word segmentation processing, Filter processing and part-of-speech processing; Acquiring a second corpus according to each document and the candidate keywords of each document; Use the TextRank model to convert the candidate keywords of each document into a keyword directed graph, and calculate the weight of the candidate keywords in each document to obtain the weighted keyword directed graph of each document; From the weighted keyword directed graph, the candidate keywords whose weights are higher than the preset weights are obtained to obtain the keyword set of each document. 4. The method according to claim 3, wherein obtaining the set of subject terms of each document in the plurality of documents comprises: Configure the calculation formula of the word influence score of the word vector according to the TextRank model. The calculation formula is as follows:

Wherein, d represents the preset damping coefficient, a represents keyword a in the weighted keyword directed graph, and Out(b) represents all keywords pointed to by keyword a in the weighted keyword directed graph In(a) represents the set of all keywords pointing to keyword a in the weighted keyword directed graph, b represents keyword b pointing to keyword a, and c represents keyword c pointing to keyword a , D(b) represents the word influence of keyword b, tfidf _a represents the product of word frequency and inverse document frequency of keyword a, tfidf _c represents the product of word frequency and inverse document frequency of keyword c, w _ba represents The weight value of keyword a and keyword b in the weighted keyword directed graph, w _bc represents the weight value of keyword b and keyword c in the weighted keyword directed graph; Performing a second process on each of the plurality of documents to obtain a third corpus and subject words in the third corpus, wherein the second process at least includes the following processes: word segmentation processing, filtering processing, extraction Initial subject word processing and statistical word frequency processing; According to the calculation formula of the word influence score, iteratively calculate the word influence score of the subject words in the third corpus, when the difference between the Nth iterative calculation and the N-1th iteration is less than the preset threshold The iterative calculation is stopped, and the subject words whose word influence score is higher than the preset word influence score are obtained from the calculation result of the Nth iterative calculation, and the subject subject set of each document is obtained. 5. The method according to claim 1, characterized in that, according to the subject word set of each document and the weighted matrix model of each document, calculate the ratio of each document in the plurality of documents to each other Target similarities between documents include: Constructing a bipartite graph model according to the subject term set of each document, to calculate the first similarity between the subject term set of each document and the subject term set of each other document; calculating the second similarity between the weighted matrix model of each document and the weighted matrix model of each other document according to the least squares distance formula of the matrix; A target similarity between each document and each other document is calculated according to the first similarity and the second similarity. 6. The method according to claim 5, characterized in that, constructing a bipartite graph model according to the subject term set of each document, to calculate the subject term set of each document and the subject term of each other document The first degree of similarity between sets includes: Constructing the bipartite graph model of each document and each other document according to the keyword set of each document; Use the Hungarian algorithm to calculate the maximum weight of bipartite matching between each document and each other document, and use the maximum weight of bipartite matching as the difference between the subject word set of each document and the subject word set of each other document first similarity. 7. The method according to claim 1, wherein obtaining target information of a plurality of customers from the document collection after the clustering comprises: According to the method of obtaining the subject word set, each clustered document set is processed, and the subject term set of each clustered document set is obtained; According to the method of obtaining the keyword set, each clustered document set is processed, and the keyword set of each clustered document set is obtained; The target information of the plurality of customers is determined according to the keyword set of each clustered document set and the keyword set of each clustered document set. 8. The method according to claim 1, wherein, according to the subject word collection of each document and the weighted matrix model of each document, calculate the relationship between each document and each other in the plurality of documents After the target similarity between documents, the method further includes: adding the target similarity between each of the multiple documents and each of the other documents to obtain a third similarity of each of the documents; Obtaining target documents whose third similarity is higher than a preset similarity from the plurality of documents to obtain a target document set; Push the target document collection to the target object. 9. A processing device for customer information, comprising: The first acquisition unit is configured to encode customer information in multiple incoming work orders of multiple customers to obtain multiple documents corresponding to the multiple incoming work orders, wherein the encoding formats of the multiple documents are the same ; a second obtaining unit, configured to obtain a weighted matrix model of each document in the plurality of documents; a third obtaining unit, configured to obtain a set of subject words of each document in the plurality of documents; A calculation unit, configured to calculate the target similarity between each of the multiple documents and each of the other documents according to the keyword set of each document and the weighted matrix model of each document; The fourth acquisition unit is configured to cluster the plurality of documents based on the PAM clustering algorithm according to the target similarity to obtain a clustered document set; The fifth obtaining unit is configured to obtain target information of multiple customers from the clustered document collection. 10. An electronic device, characterized in that it includes one or more processors and memory, and the memory is used to store one or more programs, wherein, when the one or more programs are used by the one or more When the processor is executed, the one or more processors are made to implement the customer information processing method in any one of claims 1-8.

Images